A

bot.go

@@ -449,6 +449,44 @@ func (b *Bot) sendResponse(ctx context.Context, chatID int64, text string, busin
 	return nil
 }
 
+// sendMultiResponse delivers a multi-block LLM response as separate Telegram
+// messages while keeping a single logical assistant turn in storage. The DB
+// row and chat memory hold the joined text (segments separated by blank lines),
+// so the model's next-turn context sees one assistant turn — matching today's
+// 1-reply-per-prompt invariant — even though the user saw N bubbles.
+//
+// Partial send failures (a later segment fails after earlier ones succeeded)
+// are logged but do not abort the remaining sends. The DB record is canonical:
+// the model's next turn will reference what it intended to say.
+func (b *Bot) sendMultiResponse(ctx context.Context, chatID int64, segments []string, businessConnectionID string) error {
+	if len(segments) == 0 {
+		return nil
+	}
+
+	fullText := strings.Join(segments, "\n\n")
+	if _, err := b.screenOutgoingMessage(chatID, fullText); err != nil {
+		ErrorLogger.Printf("Error storing assistant message: %v", err)
+		return err
+	}
+
+	for i, seg := range segments {
+		params := &bot.SendMessageParams{
+			ChatID: chatID,
+			Text:   seg,
+		}
+		if businessConnectionID != "" {
+			params.BusinessConnectionID = businessConnectionID
+		}
+		if _, err := b.tgBot.SendMessage(ctx, params); err != nil {
+			ErrorLogger.Printf("[%s] Error sending segment %d/%d to chat %d with BusinessConnectionID %s: %v",
+				b.config.ID, i+1, len(segments), chatID, businessConnectionID, err)
+			// Keep going: earlier segments are already in the user's chat,
+			// and the DB has the full turn recorded.
+		}
+	}
+	return nil
+}
+
 // sendStats sends the bot statistics to the specified chat.
 func (b *Bot) sendStats(ctx context.Context, chatID int64, userID int64, targetUserID int64, businessConnectionID string) {
 	// If targetUserID is 0, show global stats